Member-only story
Human thinking is very jumping, maybe we were chatting with friends in the last second, and the next second we talked about some celebrity’s gossip. It is not difficult to find that our natural dialogue is open-ended, which can start from a specific topic and end on a completely different topic. However, the open nature of real dialogue has become one of the most difficult problems in the field of machine learning.
“Google & Language Model”
Google, which started by relying on search engines, has always been fond of language. In recent years, it has made search and access more convenient through the development of machine translation, machine learning and other technologies. In 18 years, the Google team released the BERT model, which showed amazing results in the top level test of machine reading comprehension, bringing a milestone change to NLP.
Language is an important tool for understanding knowledge, making Google Translate a tool used by many people to watch content written in other languages. However, due to the limitation of the current BERT natural language model framework, the conversion result of Google translation is still not ideal. Factors include limited understanding of language content and relatively limited information that can be associated at the same time.
At the I/O conference a week…
